Sequence alignments are difficult to construct for distantly related proteins because the "correct" alignment is indistinguishable from many equally good alignments as evaluated by statistical methods. In these cases, the accuracy of an alignment can be improved by considering structural information as well as sequence similarity. In addition, information from multiple sequence alignments can provide insight into structure-function relationships when considered in conjunction with graphical visualization of protein structures. A tool that would allow interactive manipulation of sequence alignments in tandem with a computer graphics model of a protein structure facilitates this process and helps to address both needs. We have used this interface to develop an informative global alignment of the mandelate racemase superfamily using the facilities of the Computer Graphics Lab. This has provided us with a powerful new tool to investigate how enzymes are conscripted for new functions through evolutionary processes.